Generating a representative model for a plurality of models identified by similar feature data

• US 9,053,185 B1
• Filed: 04/30/2012
• Issued: 06/09/2015
• Est. Priority Date: 04/30/2012
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Liu, Jia, Bai, Yijian, Patil, Manojav, Ravichandran, Deepak, Jiampojamarn, Sittichai, Ponnekanti, Shankar
• Primary Examiner(s): Gebresilassie, Kibrom K
• Application Number:
  US13/460,469
• Time in Patent Office: 1,135 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06F 16/951 Indexing; Web crawling techniques
• CPC Class Codes: G06Q 30/0253 During e-commerce, i.e. online transactions
• CPC Class Codes: G06Q 30/0256 User search
• CPC Class Codes: G06Q 30/0277 Online advertisement

Abstract

A computerized method for generating a representative model for a plurality of different models identified by similar feature data. A processing circuit receives a plurality of different models, each model configured for use in generating a second set of network user identifiers based on a first set of network user identifiers. The processing circuit receives feature data for each of the plurality of different models, each feature data having a corresponding feature weight data. The processing circuit identifies similar models within the plurality of different models based on a similarity of the feature data between models within the plurality of different models. The processing circuit generates the representative model to represent the similar models. The representative model may be used to generate the second set of network user identifiers based on the feature data and corresponding weight data of the representative model.

Claims

1. A computerized method of generating a representative model for a plurality of different models identified by similar feature data, the method comprising:

receiving, by a processing circuit, a first model and a second model, each of the first model and the second model configured for use in identifying a second set of network user identifiers as similar to a first set of network user identifiers;

identifying, by the processing circuit, feature data associated with each of the first model and the second model, each feature data having a corresponding weight data;

selecting a network user identifier pool including a plurality of network user identifiers, a subset of the network user identifier pool including at least one network user identifier that is included in at least one of the second set of network user identifiers identified by the first model or the second model and at least one network user identifier that is not included in the at least one of the second set of network user identifiers;

determining, for each model of the first model and the second model, from the network user identifier pool, a network user identifier identified as similar to the first set of network user identifiers of the model;

determining an overlap between positive predictions and negative predictions of the first model and the second model, a positive prediction between the first model and the second model occurring when each of the first model and the second model identifies a network user identifier from the network user identifier pool as a similar network user and a negative prediction between the first model and the second model occurring when either the first model identifies a network user identifier from the network user identifier pool that is not identified by the second model or the second model identifies a network user identifier from the network user identifier pool that is not identified by the first model;

calculating, for the first model and the second model, a degree of overlap between the positive predictions and the negative predictions;

identifying, by the processing circuit, that the first model and the second model are similar responsive to determining that the degree of overlap is greater than a threshold value; and

generating, by the processing circuit, the representative model to represent the first model and the second model, the representative model configured for use in generating a second set of network user identifiers associated with the representative model based on a first set of network user identifiers associated with the representative model.

2. The method of claim 1, wherein generating the representative model comprises selecting one of the first model and the second model to be the representative model.

3. The method of claim 1, wherein generating the representative model comprises generating the representative model by including feature data from a pool of the feature data associated with each of the first model and the second model and calculating an average corresponding weight data for the each feature data.

4. The method of claim 1, wherein generating the representative model comprises generating the representative model by ranking the feature data and the corresponding weight data associated with each of the first model and the second model and selecting a predetermined number of high-ranking feature data from each of the first model and the second model.

5. The method of claim 1, further comprising providing the second set of network user identifiers associated with the representative model from the first set of network user identifiers associated with the representative model, based on the feature data and the corresponding weight data of the representative model.

6. A system of generating a representative model for a plurality of different models identified by similar feature data, the system comprising:

a memory; and

one or more processors, the processors configured to

receive a first model and a second model, each of the first model and the second model configured for use in identifying a second set of network user identifiers as similar to a first set of network user identifiers;

receive feature data associated with each of the first model and the second model, each feature data having a corresponding weight data;

select a network user identifier pool including a plurality of network user identifiers, a subset of the network user identifier pool including at least one network user identifier that is included in at least one of the second set of network user identifiers identified by at least one of the first model or the second model plurality of different models and at least one network user identifier that is not included in the at least one of the second set of network user identifiers identified by at least one of the plurality of different models;

determine, for each model of the first model and the second model, from the network user identifier pool, a network user identifier identified as similar to the first set of network user identifiers of the model;

determining an overlap between positive predictions and negative predictions of the first model and the second model, a positive prediction between the first model and the second model occurring when each of the first model and the second model identifies a network user identifier from the network user identifier pool as a similar network user and a negative prediction between the first model and the second model occurring when either the first model identifies a network user identifier from the network user identifier pool that is not identified by the second model or the second model identifies a network user identifier from the network user identifier pool that is not identified by the first model; and

calculate, for the first model and the second model, a degree of overlap between the positive predictions and the negative predictions;

identify that the first model and the second model are similar responsive to determining that the degree of overlap is greater than a threshold value; and

generate the representative model to represent the first model and the second model, the representative model configured for use in generating a second set of network user identifiers associated with the representative model based on a first set of network user identifiers associated with the representative model.

7. The system of claim 6, wherein generating the representative model comprises selecting one of the first model and the second model to be the representative model.

8. The system of claim 6, wherein generating the representative model comprises generating the representative model by including the feature data from a pool of the feature data associated with each of the first model and the second model and calculating an average corresponding weight data for the each feature data.

9. The system of claim 6, wherein generating the representative model comprises generating the representative model by ranking the feature data and the corresponding weight data associated with each of the first model and the second model and selecting a predetermined number of high-ranking feature data from each of the first model and the second model.

10. The system of claim 6, further comprising providing a second set of network user identifiers associated with the representative model from a first set of network user identifiers associated with the representative model, based on the feature data and the corresponding weight data of the representative model.

11. The system of claim 10, further comprising generating display data configured to display the second set of user identifiers associated with the representative model on a user interface.

12. A non-transitory computer-readable medium having instructions thereon that cause one or more processors to perform operations, the operations comprising:

receiving a first model and a second model, each of the first model and the second model configured for use in identifying a second set of network user identifiers as similar to a first set of network user identifiers;

receiving feature data associated with each of the first model and the second model, each feature data having a corresponding weight data;

selecting a network user identifier pool including a plurality of network user identifiers, a subset of the network user identifier pool including at least one network user identifier that is included in at least one of the second set of network user identifiers identified by the first model or the second model and at least one network user identifier that is not included in the at least one of the second set of network user identifiers;

determining, for each model of the first model and the second model, from the network user identifier pool, a network user identifier identified as similar to the first set of network user identifiers of the model;

determining an overlap between positive predictions and negative predictions of the first model and the second model, a positive prediction between the first model and the second model occurring when each of the first model and the second model identifies a network user identifier from the network user identifier pool as a similar network user and a negative prediction between the first model and the second model occurring when either the first model identifies a network user identifier from the network user identifier pool that is not identified by the second model or the second model identifies a network user identifier from the network user identifier pool that is not identified by the first model;

calculating, for the first model and the second model, a degree of overlap between the positive predictions and the negative predictions;

identifying, by the processing circuit, that the first model and the second model are similar responsive to determining that the degree of overlap is greater than a threshold value; and

generating the representative model to represent the first model and the second model, the representative model configured for use in generating a second set of network user identifiers associated with the representative model based on a first set of network user identifiers associated with the representative model.

13. The computer-readable medium of claim 12, wherein the operations further comprise providing the second set of network user identifiers associated with the representative model from the first set of network user identifiers associated with the representative model, based on the feature data and the corresponding weight data of the representative model.